Comparing strategies of genomic selection to increase oil palm fresh fruit bunch yield

(1)

Comparing strategies of genomic selection

p

g

to increase oil palm fresh fruit bunch yield

p

y

D id CROS

1

_{M i DENIS}

1

_J

_M

_BOUVET

1

_Lé

_{ld SANCHEZ}

2

David CROS

1

_{, Marie DENIS}

1

_{, Jean-Marc BOUVET}

1

_{, Léopoldo SANCHEZ}

2

David CROS , Marie DENIS , Jean Marc BOUVET , Léopoldo SANCHEZ

1 _{Genetic improvement and adaptation of Mediterranean and tropical plants} 1 _{Genetic improvement and adaptation of Mediterranean and tropical plants}

Research Unit (AGAP), CIRAD, 34398 Montpellier, France( ) p

2 _{Forest Tree Improvement, Genetics and Physiology Research Unit}_p _, _y _gy

(AGPF) INRA 45075 Orléans France (AGPF), INRA, 45075 Orléans, France

Introduction

Genomic selection (GS) can increase hybrid performance in Genomic selection (GS) can increase hybrid performance in

b d i di id l H t i i l t it lt f

crossbred individuals. Heterosis in a complex trait can result fromp multiplicative interaction between more simple and additive multiplicative interaction between more simple and additive

t I il l th b h d ti i th d t f

components. In oil palm, the bunch production is the product ofp p p p bunch weight (ABW) and bunch number (BN) two additive and bunch weight (ABW) and bunch number (BN), two additive and

ti l l t d t it Oil l b di li i l

negatively correlated traits. Oil palm breeding relies on reciprocalg y p g p recurrent selection (RRS) between two complementary populations recurrent selection (RRS) between two complementary populations

f ABW d BN ll D li d Af i l ti lik L Mé

for ABW and BN, usually Deli and an African population like La Mé.

This study aimed at evaluating the potential of reciprocal This study aimed at evaluating the potential of reciprocal

recurrent genomic selection (RRGS) as an alternative tog ( )

_Deli

_{La Mé}

conventional RRS

Deli

La Mé

conventional RRS.

Fig 1 Simulation of the initial breeding populations (generation 0) Fig. 1 Simulation of the initial breeding populations (generation 0).

R d ti ll d hi t ti d ift ilib i ft 2400 ti N t l Random mating allowed reaching mutation-drift equilibrium after 2400 generations. Natural

Material and methods

selection was applied to increase bunch weight (ABW) in population A (in blue) and bunch

Material and methods

pp g ( ) p p ( )

number (BN) in population B (in red) Bottleneck events were at the origin of Deli and La Mé

We simulated two realistic oil palm breeding populations (Fig 1) number (BN) in population B (in red). Bottleneck events were at the origin of Deli and La Mé _{populations In subsequent generations artificial selection (mass selection and RRS) was} We simulated two realistic oil palm breeding populations (Fig. 1)

d d f ti RRGS ith RRS (Fi 2) populations. In subsequent generations, artificial selection (mass selection and RRS) was ( f ) Q f

and compared over four generations RRGS with RRS (Fig. 2). applied to increase bunch production (FFB, which is the product of ABW by BN). QTL for

The goal of all breeding strategies was to select the best

pp p ( p y )

ABW and BN were assigned after the first 2400 generations of random mating.

The goal of all breeding strategies was to select the best

i di id l i th 2 t l l ti f h b id f

ABW and BN were assigned after the first 2400 generations of random mating. RRS: reciprocal recurrent selection

individuals in the 2 parental populations for hybrid performance on RRS: reciprocal recurrent selection bunch production For RRGS we used 2500 SNP and the

bunch production. For RRGS, we used 2500 SNP and the phenot pes of h brids as data records in the GBLUP method to

phenotypes of hybrids as data records in the GBLUP method to _{Fig 2 Simulation of the}

obtain the parental genomic estimated breeding values We studied Fig. 2 Simulation of the _{breeding strategies applied} obtain the parental genomic estimated breeding values. We studied

the effects of 4 parameters on the selection response in hybrids: breeding strategies applied

the effects of 4 parameters on the selection response in hybrids: _{for 4 generations to the}

(1) the molecular data used to calibrate the GS model: in initial breeding populations.g

(1) the molecular data used to calibrate the GS model: in RRGS PAR we only used parental genotypes and in RRGS HYB

initial breeding populations.

The progeny tests (in purple)

RRGS_PAR, we only used parental genotypes and in RRGS_HYB The progeny-tests (in purple)

l t d 120 i di id l

we also used genotypes of hybrid individuals, taking into account evaluated 120 individuals per

we also used genotypes of hybrid individuals, taking into account

the parental origin of marker alleles; (2) the frequency of calibration population.

the parental origin of marker alleles; (2) the frequency of calibration popu a o

RRGS: reciprocal recurrent

of GS model [every generation, every 2 or every 4 generations]; (3) RRGS: reciprocal recurrent _i _{l ti}

of GS model [every generation, every 2 or every 4 generations]; (3)

for RRGS HYB the number of genotyped hybrids [300 1000 genomic selection

for RRGS_HYB, the number of genotyped hybrids [300, 1000, _{(RRGS PAR when parental}

1700 individuals] and (4) the number of selection candidates [120,] ( ) [ , (_{genotypes were used to}_ p

300 individuals] genotypes were used to calibrate GS model and

300 individuals]. calibrate GS model and

RRGS HYB h h b id RRGS_HYB when hybrid

gametotypes were also used).

Results and discussion

gametotypes were also used). RRGS was used to increase

Results and discussion

RRGS was used to increase

l ti i t it d /

RRGS PAR

RRGS HYB

With RRS the annual selection response in hybrid bunch

RRGS_PAR

RRGS_HYB

selection intensity and / or

With RRS the annual selection response in hybrid bunch

d ti 0 30% (Fi 3) Th hi h t l l ti shorten generation interval

production was 0.30% (Fig. 3). The highest annual selection g

response was made by RRGS HYB with progeny-tests every four response was made by RRGS_HYB with progeny tests every four

ti 300 did t d t l t 1 000 t d h b id

RRS

generations, 300 candidates and at least 1,000 genotyped hybrids.

RRS

The annual response was 0.49%, ie almost two thirds higher The annual response was 0.49%, ie almost two thirds higher than ann al response of RRS (P<0 01)

than annual response of RRS (P<0.01).

With RRGS HYB the annual selection response was 0 06% With RRGS_HYB, the annual selection response was 0.06% higher when the selection was made among 300 individuals thang g 120 individuals (P<0 001) The annual selection response was 120 individuals (P<0.001). The annual selection response was

i il i h 2 d 4 i b

similar with progeny-tests every 2 and every 4 generations, butp g y y y g , 0 10% higher than with progeny-tests every generation (P<0 001) 0.10% higher than with progeny-tests every generation (P<0.001).

Th l l ti i il ith 1 000 d 1 700

The annual selection response was similar with 1,000 and 1,700p , , genotyped hybrids but 0 13% higher than with 300 genotyped genotyped hybrids, but 0.13% higher than with 300 genotyped h b id (P 0 001)

hybrids (P<0.001).y ( )

With RRGS PAR the annual selection response was 0 05% higher With RRGS_PAR, the annual selection response was 0.05% higher when the selection was made among 300 individuals than 120 when the selection was made among 300 individuals than 120 individuals (P<0 01) The annual selection response was similar individuals (P<0.01). The annual selection response was similar with progeny-tests every 2 and every 4 generations, but 0.07% with progeny tests every 2 and every 4 generations, but 0.07% higher than with progeny tests every generation (P<0 01)

higher than with progeny-tests every generation (P<0.01).

Conclusion

Both RRGS HYB and RRGS PAR could lead to a much higher _ _ g selection response for FFB in hybrids than RRS because they selection response for FFB in hybrids than RRS, because they allowed reducing the generation interval and increasing g g g

selection intensity selection intensity.

The best strategy was RRGS HYB with progeny-tests every 4 gy _ p g y y

generations 300 candidates and genotyping at least 1 000 hybrids _{Fig 3 Ann al response to selection after fo r generations of selection (in percentage} generations, 300 candidates and genotyping at least 1,000 hybrids.

H RRGS PAR i h 2 4 Fig. 3 Annual response to selection after four generations of selection (in percentage However, RRGS PAR with progeny-tests every 2 or every 4 , _ p g y y y of bunch production of hybrids in generation 0).

generations and 300 selection candidates was an interesting (and

p y g )

For RRGS HYB 1 700 hybrids were used to calibrate the GS model

generations and 300 selection candidates was an interesting (and

h ) lt ti

For RRGS_HYB 1,700 hybrids were used to calibrate the GS model.

GGGG: GS model calibrated every generation (with progeny tests) GMGM: GS model

cheaper) alternative. GGGG: GS model calibrated every generation (with progeny-tests), GMGM: GS model

lib t d 2 ti d l ti d l k i th th ti

p )

calibrated every 2 generations and selection made only on markers in the other generations, GMMM: GS model calibrated every 4 generations.

GMMM: GS model calibrated every 4 generations.

120 / 300: number of candidates per generation and population (in RRS the candidates were 120 / 300: number of candidates per generation and population (in RRS the candidates were th 120 t t d i di id l t l l ti )

the 120 progeny-tested individuals per parental population).

Values were means over 5 replicates. Values with the same letter were not significantly p g y different at P=0 01